摘要：The image blocks to be compressed are classified according to their activity degree before
presenting them to Kohonen’s network for a vector quantization. This classification allows
discrimination between the blocks with high activity (edge blocks) and blocks with law activity. The
blocks of high activity are divided into small blocks of 2x2 pixels. Blocks of high and law activity are
coded separately with different codebooks. We have obtained a noticeable improvement of visual
quality of all the rebuild medical images while keeping an important compression rate.
